A new AI model developed by researchers at the New Jersey Institute of Technology (NJIT), Princeton University, and NASA’s Ames Research Centre can detect subtle signs of emerging solar active regions before they become visible on the Sun.
Called EarlyDetect, the model analyses changes in the Sun’s acoustic waves and magnetic field using observations from NASA’s Solar Dynamics Observatory (SDO).
In tests, it identified the signals associated with new active regions an average of 9.24 hours before they appeared at the surface.
The result could eventually improve forecasts of solar activity by providing earlier warning of regions capable of producing solar eruptions.
However, the researchers stress that EarlyDetect is still experimental and cannot yet predict whether an active region will produce a flare or coronal mass ejection.
Detecting solar activity beneath the surface
Solar active regions are areas of intense magnetic activity where sunspots develop. They can emerge over several hours, though the structures may take days to develop fully.
The challenge for researchers is that their formation begins below the Sun’s visible surface. Researchers cannot observe the magnetic structures directly as they rise through the solar interior, but they can leave faint traces in the Sun’s acoustic activity.
EarlyDetect was designed to identify those traces. The system processes hourly acoustic power maps alongside magnetic-field measurements collected by the Helioseismic and Magnetic Imager aboard SDO.
The acoustic maps are generated from observations of solar vibrations recorded every 45 seconds. These measurements let researchers examine subtle changes in waves moving through the Sun and look for patterns linked to emerging magnetic regions.
AI finds signals conventional methods can miss
The research team used a transformer-based architecture, a form of machine learning also used in modern language models. Instead of analysing words and sentences, EarlyDetect learns patterns within sequences of solar observations.
One unexpected finding involved data filtering. Researchers initially applied a filtering method intended to remove irrelevant fluctuations and make useful patterns easier for the model to identify. Instead, the approach reduced its forecasting performance.
The discarded fluctuations contained information about the earliest stages of active-region formation. Removing them effectively erased some of the weak signals the AI needed to predict early.
After training on SDO/HMI observations, the researchers evaluated EarlyDetect on active regions not included in its training data.
The strongest version of the model detected precursor signals 9.24 hours before the regions became visible, outperforming a conventional transformer and an earlier benchmark method.
What earlier solar warnings could mean
Earlier detection of emerging active regions could eventually help space-weather forecasting.
Solar active regions can become sources of powerful flares and coronal mass ejections, which can disturb Earth’s space environment. Major events can interfere with satellite operations, radio communications and electrical infrastructure.
An additional several hours of warning for solar eruptions could therefore give satellite operators, communications providers and grid operators more time to assess potential risks and prepare for possible disruption.
But detecting an active region is not the same as forecasting a major solar storm. Many emerging regions do not generate significant flares or coronal mass ejections, so further forecasting is needed to determine whether an identified region is likely to become hazardous.
A new dataset for solar forecasting
The researchers released the Solar Active Region Emergence Dataset, or SolARED, to let other scientists study the problem and test new forecasting techniques.
An accompanying Solar Active Region Portal provides an interactive way to explore the observations.
Making the dataset publicly available could help broaden research into machine-learning approaches for solar forecasting, particularly by giving researchers a common set of observations against which different models can be evaluated.
For now, EarlyDetect remains a research prototype. It can produce false alarms and sometimes identify an emerging region too late.
The team says it will need to test the system against a much larger number of solar events before it can be considered suitable for operational forecasting.
Even so, the findings suggest that the earliest stages of solar activity may be detectable well before sunspots appear, offering a potential new route towards earlier warnings of the conditions that can ultimately lead to solar eruptions.
